A Hybrid Approach to Detect Botnet Attacks in IoT Environments using Machine Learning

The use of IoT devices has greatly raised cyber security threats, especially botnet attacks since it affect services and data. To overcome these difficulties, it proposes a hybrid model for IoT botnet detection scene including Autoencoder to detect the anomaly and LSTM for temporal-spatial analysis. To improve the reliability in detection an effective Bayesian decision fusion mechanism is used for the fusion of decisions. The detection experiment of the proposed model is tested on benchmark datasets with 98.7% and 97.8% accuracy, F1-score of 96.8% and 96.1%, and false-positive ratios of 1.5% and 1.8%. Furthermore, the model projected a low latency of 35 ms, which makes it suitable for real-time applications. It is evident that the proposed approach results in better detection performance compared to other conventional methods in addition to less computation time. These results indicate that the proposed framework can be implemented effectively to detect and mitigate botnet attacks in an IoT environment that has limited resources.